Overview
- Anthropic announced Claude Opus 4.8 on May 28 and partners made it broadly available within days on AWS Bedrock, GitHub Copilot, Snowflake Cortex AI public preview, Microsoft Foundry, and Google Vertex AI.
- Opus 4.8 introduces Dynamic Workflows for running many parallel subagents on large coding tasks, an Effort Control slider to trade reasoning depth for latency, and claims of reduced false confidence in code generation.
- Developers can access Opus 4.8 via Anthropic’s API and major clouds and IDE integrations, with a 1 million token input context and faster, lower-cost fast modes aimed at real-world coding and agentic workflows.
- Independent benchmarks such as ITBench-AA show frontier models underperforming on many agentic IT tasks, so companies are advised to run scoped tests, restrict permissions, add audit logs, and keep human oversight before full production use.
- The launch fits Anthropic’s broader expansion that includes larger funding and multi-hyperscaler compute deals, which speeds enterprise placement but raises supplier concentration and governance questions for production agents.